CUDA out of memory

[yitengfei120011]

adata = model.fit(lr=lr, epochs=epochs,)

The CUDA memory is insufficient. How can we reduce its usage?

OutOfMemoryError: CUDA out of memory. Tried to allocate 222.41 GiB (GPU 0; 23.56 GiB total capacity; 583.94 MiB already allocated; 22.66 GiB free; 612.00 MiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.

[ProDong0512]

Sorry for the late reply. scNiche is compatible with large datasets using a batch training strategy.

You may decrease the scale of batch_size to avoid generating a huge graph that eats up your CUDA memory. Practically, you can adjust the parameter batch_num in the step 'prepare_data_batch' to a larger number, so that the size of per batch would be smaller.

You may see it in tutorial_DLPFC.ipynb or tutorial_MERFISH.ipynb.